Data Analysis Summary

1. Data Analysis Summary

2. Elephant in the room

3. General Comments • General understanding that informatics is integral in medical sequencing and other –omics in clinical settings • About 80% of attendees were actively involved in data analysis • Clinical practitioners also present • Talking about data analysis is difficult • We do not yet have language to do so • Complicated • Not clear what details are important

4. Overview • Analytical Validity • Enhance Clinical Utility • Data Sharing • Messages to NCI

5. Analytical Validity • Definition of “mutation”, both at the level of variant calling and with “established” calls, remains unclear • Reproducible analytical methods in both research and clinical practice needed • Versioning of raw data and processed data • Versioning of data analysis pipelines • Versioning of auxiliary data (gene models, sequences, etc.)

6. Analytical Validity • Understanding that algorithms that use all the information (tumor/normal or multiple tumor samples) yields higher sensitivity and specificity • Data archiving and sharing becomes important • Archiving blocks is not enough—best to archive data as well • Need to provide confidence associated with variants since no test is 100% accurate

7. Analytical Validity • “Reference” genome • A computational reference is important to allow communication of findings • Lack of adequate ground-truth datasets precludes rigorous evaluation of analysis, particularly in quantifying false-negative rate • Tumor heterogeneity, tissue heterogeneity, and even stochastic sampling at the sample level remain challenges in establishing analytical validity

8. Enhance Clinical Utilty • Lacking large (10s of thousands of patients), well-annotated databases of normal or disease patients • Definition of clinically actionable remains unclear • Reference database of clinically actionable variants • Does not exist • Will be challenging to update and maintain • Incorporating clinical context is difficult but probably necessary if one is to truly achieve precision medicine

9. Enhance Clinical Utility • Establish methods of reporting that empower the clinician • Enough detail to be helpful • Not so much detail as to be unintelligible • Integrate with online databases and knowledge

10. Data Sharing • Need to establish standards for data sharing in both research and clinical venues (think Myriad and BRCA1 testing) • Protocols, both computational and laboratory • Controlled vocabularies • Clinical data • The data themselves • Consider incentivizing data sharing • Pay-to-play sharing • NCI mandate

11. Data Sharing • What constitutes de-identified data? • Need to respect rights of patients, including protecting AND sharing data • Need some way to feed clinical information back to into the informatics pipeline • Clinicians need actionable information with as much interpretation with regard to literature and knowledgebase as possible

12. Messages to NCI • Critical need to establish ground truth datasets and biologics • Fix TCGA! • NCI should collect and maintain knowledgebase of “clinically actionable” information (variants, genes, pathways) • Start by collecting and updating lists from large medical centers • Enhance PDQ database to include computable information on molecular targets under study

13. Messages to NCI • More input is needed when NCI is planning bioinformatics, computational biology, and biomedical informatics • Granting mechanism? • Less top-down approach to informatic • Establish and ENFORCE rational data sharing mechanisms for NCI-sponsored clinical trials • SRA is not the answer….

14. Patient and Population Characteristics Gene Expression Gene Copy Number Transcriptional Regulation phenotype DNA Methylation Chromatin Structure and Function Sequence Variation

15. Questions